Why Should the RealTheory Agent Run with Only One Replica?

Prev Next

Background

The RealTheory Agent (also referred to as the Collector) is deployed as a single-replica Deployment inside your Kubernetes cluster. It is designed to collect metrics, performance, cost, and health data from the entire cluster and stream that data to the RealTheory platform.

The Agent is architected to operate as a single logical collector per cluster. It does not require horizontal scaling for performance or availability.

Expected Deployment Configuration

By default, the RealTheory Agent is deployed with:

  • replicas: 1
  • A single active pod per cluster

This configuration is sufficient for clusters of all supported sizes.

What Happens When You Increase the Replica Count

If you manually increase the replica count of the RealTheory Agent Deployment (for example, setting replicas: 2 or higher), each replica operates independently and performs the same collection and transmission tasks.

This results in:

  • Duplicate metric collection
  • Duplicate data transmission to the RealTheory backend
  • Increased outbound network bandwidth usage
  • Increased CPU and memory consumption within the cluster

The impact scales linearly:

  • 2 replicas ≈ 2× bandwidth and resource consumption
  • 3 replicas ≈ 3× bandwidth and resource consumption
  • 4 replicas ≈ 4× bandwidth and resource consumption

Running multiple replicas does not improve data collection accuracy, availability, or performance.

When to Investigate

If you increased the replica count because you were experiencing:

  • Data gaps
  • Excessive restart activity (more than a few times each week)
  • Perceived performance issues

We recommend reviewing:

  • Pod health and restart status
  • Node resource availability
  • Network connectivity to RealTheory services

If concerns persist, please contact RealTheory Support so we can review logs and metrics with you.

Solution

Ensure that the RealTheory Agent Deployment is configured with:

replicas: 1

If multiple replicas are currently running, scale the Deployment back to a single replica to prevent unnecessary resource and bandwidth consumption.